Integrated data anlaysis via clustering

Abstract

Discovering the exact activities of a compound is important in drug discovery. The compound may or may not have the desired effects and indeed, if a compound reacts to an off-target and this is not seen, severe side effects can be the result. A single source of information is limited by its specific point of view. The integration of multiple data sources can offer more insight into the mechanism of action and help to shine a light on the global picture of the working of compounds. Several integrative data clustering techniques were performed on two data sets which were accompanied by two data sources each. By comparing the clustering on the separate sources with the integrative analyses, the influence of each could be investigated. A best integrative method is not declared. Rather interest lies in clusters that are found to be stable over the different methods. These compounds are indicated to be similar on different aspects of the underlying biology. It can than be hypothesized that the data sources are related for those compounds. If compounds do not show a clear resemblance to one or possibly multiple groups, they can be clustered differently for each method. It was seen that the differential expression of the cluster is greatly influenced by the compounds joined to them.